Analysis of Human Faces using a Measurement-Based Skin Reflectance Model

Abstract
we have measured 3D face geometry, skin reflectance, and subsurface scattering using custom-built devices for 149 subjects of varying age, gender, and race. we developed a novel skin reflectance model whose parameters can be estimated from measurements. the model decomposes 分解 the large amount of measured skin data into a spatially-varying analytic BRDF, a diffuse albedo map, and diffuse subsurface scattering. our model is intuitive, physically plausible, and——since we do not use the original measured data——easy to edit as well. high-quality renderings come close to reproducing real photographs. the analysis of the model parameters for our sample population reveas variations according to subject age, gender, skin type, and external factors (e.g., sweat, cold, or makeup). using our statistics, a use can edit the overall appearance of a face (e.g., change skin type and age) or change small-scale features using texture synthesis (e.g, adding moles and freckles). we are making the collected statistics publicly available to the research community for applications in face synthesis and analysis.

key worlds: face modeling, reflection models, data-driven models

1 introduction
one of the most difficult computer graphics challenge is creating digital faces that are indistinguishable from real ones. the process of capturing an actor’s performance and likeness has accurately been named “Digital face cloning”. digital face cloning has many applications in movies, games, medicine, cosmetics, computer vision, biometrics, and virtual reality. despite recent progress, however, it remains a challenging problem because humans are incredibly sensitive to the subtleties of real faces. 然而，尽管最近取得了进展，这仍然是一个具有挑战性的问题，因为人类对真实面孔的微妙之处极为敏感。

in this paper, we focus on modeling skin reflectance of human faces, an important aspect of face appearance. our primary goal is to accurately capture and reproduce the face reflectance of a person. we use custom-built devices to measure 3D face geometry, light reflection, and subsurface scattering. our reflectance modle consists of a spatially-varying analytic reflectance model (Torrance-Sparrow or Blinn-Phong), a diffuse albedo map, and diffuse scattering coeficients. we concisely 简明地 describe the mathematics of the model and show how its components can be robustly estimated from measured data. the model is compact, intuitive, and easy to use. images generated from it come very close to real photographs (see figure 1).

a second goal is to capture and describe the variation of face reflectance for a sample population. we measured the parameters of our reflectance model for 149 people of different race, gender, and age. the result is a unique database of spatially-varying parameters for the Torrance-Sparrow and Blinn-Phong reflectance models, histrograms of skin albedo, and diffuse scattering parameters for faces. from this data we computed the distributions of the parameters (most of them are Gaussian) and analyzed them using statistical methods. we publish the key findings to our analysis in this paper, and make the complete statistical data available to the research community. to our knowledge, this is the first study of face reflectance of this kind.

the results of our analysis have applications in realistic face synthesis, face reconstruction from photographs, inverse rendering, and computer vision. we also demonstrate how an artist can edit faces using the parameter distributions. for example, we can change the skin type of a model by transfering statistics of albedo and reflectance parameters (see figure 1). we can also use texture synthesis/analysis to transfer small-scale skin features, such as moles 痣 and freckles雀斑. the combination of these approaches allows an artist to convincingly tranfer face appearance between individuals.

2 previous work
properties of human skin have been measured and studied extensively in the computer graphics, biomedical, and computer vision communities. an excellent survey on the physiological and anatomical properties of skin and the state of the art in skin appearance modeling has been published by Igarashi et al. [2005]. in this section we provide an overview of the relevant work in the area of face appearance modeling in computer graphics.

analytic face appearance models: the simplest reflectance model for skin is the Bidirectional Reflectance Distribution Functoin (BRDF) [Nicodemus et al. 1977]. Dana et al. [1999] measured the BRDF of skin and made the data available in the Columbia-Utrecht Reflectance and Texture (CuReT) database (http://www.cs.columbia.edu/CAVE/curet/). because the data was not collected in vivo 活体组织 it does not accurately reflect the properties of living skin. Marschner et al. [1999;2000b] measured spatially uniform BRDFs on the foreheads 额头 of four subjects of different gender, race, and age. they observed that the BRDF of skin exhibits strong forward scattering in off-specular direction at grazing angles. 他们观察到皮肤的BRDF在掠射角的非镜面方向上呈现强烈的前向散射